前段时间遇到HDFS的各种超时,这里总结下遇到的超时:
IPC
1.IPC connect timeout:
在Client端setupConnection设置,2.0.3之前是20s,之后是可配置ipc.client.connect.timeout
2.IPC max idle time:ipc.client.connection.maxidletime设置,默认20s,若在20s内calls依然为空则主动关闭;在Server端该值为×2
3.IPC ping timeout:在Client端由ipc.ping.interval设置,默认1min,定期刷新最后更新时间.
4.RPC timeout:在call调用内由客户端设置
SOCKET
1.IPC 连接的Socket timeout:若rpctimeout>0则去rpctime,否则为ipc.ping.interval
2.Connect/Read timeout:dfs.client.socket-timeout,默认READ_TIMEOUT = 60 * 1000,READ_TIMEOUT_EXTENSION= 5 * 1000,connect timeout还未分离(HDFS-1075)
3.Write timeout:dfs.datanode.socket.write.timeout,默认WRITE_TIMEOUT = 8 * 60 * 1000,WRITE_TIMEOUT_EXTENSION= 5 * 1000
本文转自MIKE老毕 51CTO博客,原文链接:http://blog.51cto.com/boylook/1367248,如需转载请自行联系原作者